OpenAI has announced the launch of the new language model GPT-5.5, which, according to the company, is the most intelligent and intuitive in its lineup. Developers emphasize that GPT-5.5 better understands user intentions and performs tasks with fewer tokens, allowing for increased efficiency in everyday work.
This is reported by Finway
Expanded Capabilities and Tool Integration
GPT-5.5 expands functional capabilities for task automation: the model can write and debug code, analyze data, create documents, and manage various tools directly from a single interface. OpenAI considers this release an important step towards unifying ChatGPT, Codex, and AI Browser into a single service — the so-called “super app,” which will enable the resolution of most work tasks within a single ecosystem.
“The new version also brings closer the launch of the ‘super app’ — a single service that aims to combine ChatGPT, Codex, and AI Browser. OpenAI expects that this integration will allow for the completion of an increasing number of work tasks within one ecosystem.”
The model is already available to Plus, Pro, Business, and Enterprise subscription users in ChatGPT and Codex, as well as in the GPT-5.5 Pro version for corporate subscriptions. Access to the API is promised to be opened soon, with a base cost of $5 for 1 million input tokens and $30 for 1 million output tokens, and for the Pro version — $30 and $180 respectively.
Impressive Results in Programming, Science, and Security
Particular attention in GPT-5.5 is given to programming. In the Terminal-Bench 2.0 tests, the model scored 82.7%, and in the SWE-Bench Pro — 58.6%. An internal benchmark, Expert-SWE, showed that GPT-5.5 outperforms the previous version GPT-5.4 in performing complex engineering tasks with long-term planning, achieving this with lower token costs.
In the GDPval benchmark, which assesses professional intellectual work in 44 fields, GPT-5.5 achieved 84.9%, while in OSWorld-Verified, which tests performance in real computer environments, it scored 78.7%. In complex customer service scenarios (Tau2-bench Telecom), the model demonstrated 98% without the need for additional tuning. OpenAI also notes high results for GPT-5.5 in financial analysis, modeling, and office tasks.
In scientific tests, such as GeneBench, which evaluates data analysis in genetics and quantitative biology, GPT-5.5 showed significant growth compared to the previous version. In the BixBench benchmark for bioinformatics, the model achieved the best result among published systems. The company believes that GPT-5.5 is already capable of accelerating real scientific research.
OpenAI has also focused on security: GPT-5.5 has strengthened control over dangerous queries, added new risk classifiers, and implemented additional measures against abuse. The company’s experts assess the level of cybersecurity, as well as safety in biological and chemical fields, as high.